Deepseek + Claude

🚀 Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP

An MCP server that combines the inference capabilities of DeepSeek R1 and the response generation capabilities of Claude 3.5 Sonnet, implemented via OpenRouter. This implementation uses a two-stage process where DeepSeek performs the initial inference (with a context limit of 50,000 characters), and then the inference results are injected into Claude to generate the final response (with a context limit of 600,000 characters).

🚀 Quick Start

This project is an MCP server that integrates the inference ability of DeepSeek R1 and the response generation ability of Claude 3.5 Sonnet through OpenRouter. It operates in a two - stage process to provide efficient and accurate responses.

✨ Features

Two - stage Processing

• Use DeepSeek R1 for preliminary inference (with a 50k character context).
• Use Claude 3.5 Sonnet to generate the final response (with a 600k character context).
• Access both models through OpenRouter's unified API.
• Inject DeepSeek's inference results into Claude's context.

Intelligent Conversation Management

• Detect active conversations based on file modification times.
• Handle multiple concurrent conversations.
• Automatically filter out ended conversations.
• Support context clearing when needed.

Optimized Parameters

• Model - specific context limits:
  • DeepSeek: 50,000 characters for focused inference.
  • Claude: 600,000 characters for comprehensive response generation.
• Recommended settings:
  • temperature: 0.7 (for creativity and balance).
  • top_p: 1.0 (for a full probability distribution).
  • repetition_penalty: 1.0 (to prevent repetition).

📦 Installation

Installation via Smithery

Automatically install DeepSeek Thinking and Claude 3.5 Sonnet through Smithery:

npx -y @smithery/cli install @newideas99/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP --client claude

Manual Installation

1. Clone the repository:

git clone https://github.com/yourusername/Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP.git
cd Deepseek-Thinking-Claude-3.5-Sonnet-CLINE-MCP

2. Create and activate a virtual environment (recommended):

python -m venv .venv
source .venv/bin/activate  # For Linux/Mac
# Or
.\.venv\Scripts\activate  # For Windows

3. Install dependencies:

pip install -r requirements.txt

💻 Usage Examples

Basic Usage

To start the MCP server, run:

python server.py

Then, you can use any HTTP client (such as curl or Postman) to send requests to http://localhost:5000.

Advanced Usage

For automatic reconstruction in the development environment:

npm run watch

🔧 Technical Details

How it Works

1. Inference Stage (DeepSeek R1)

   • Use OpenRouter's inference token feature.
   • Modify the prompt to output 'completed' while capturing the inference process.
   • Extract the inference results from the response metadata.

2. Response Stage (Claude 3.5 Sonnet)

   • Receive the original prompt and DeepSeek's inference results.
   • Generate the final response by combining the inference results.
   • Maintain the conversation context and history.

📄 License

This project is under the MIT License. For detailed information, please refer to the LICENSE file.

🤝 Contributors

This implementation is based on the RAT (Retrieval - Augmented Thinking) concept proposed by Skirano. This approach enhances AI responses through structured inference and knowledge retrieval. Specifically, it combines the inference capabilities of DeepSeek R1 with the response generation capabilities of Claude 3.5 Sonnet via OpenRouter's unified API.